Nobody can say if androids will dream of electric sheep, although, according to new research from Los Alamos National Laboratory, they will need periods of rest that offer benefits similar to the ones that sleep offer to live brains.Content: digital intelligent brains tend to nap
We are studying spiking neural networks, systems which learn much like living brains, said Yijing Watkins, a Los Alamos National Computer Laboratory scientist. We fascinated by the opportunity to train a neuromorphic processor in a way similar to the way people and other biological systems learn from their environment during their childhood.
After continuous unattended testing, Watkins and her study team considered the network simulations dysfunctional. When the networks have been exposed to states analogous to the waves experienced in sleep by living brains, stability has been restored. “It was like we offered a night of rest to the neural networks,” Watkins said.
Digital intelligent brains tend to nap
This finding has been made whilst the development team has been designing neural networks that are strongly linked to human beings and other biological systems. Initially, the community struggled with the stabilization and classification of artificial neural networks following unpredictable dictionary training without any prior examples to evaluate them.
When we try to use biologically realistic, spiking neuromorphic processors or when it comes to understanding biology itself, the problem of how systems can become unstable is only posed, says computer scientist Los Alamos and co-author Garrett Kenyon.
This question rarely appears to the overwhelming majority of machine learning, deep science, and AI investigators, as they research the privilege of carrying out global mathematical operations in the artificial systems that control the overall system’s dynamical advantage.
Nap and Gaussian noise
Gaussian noise, named after Carl Friedrich Gauss, is statistical noise having a probability density function equal to that of the normal distribution.
The investigators describe the alternative of opening the networks to an artificial sleep equivalent as more a last move toward stabilization. Different forms of noise, approximately equivalent to that heard between stations when tuning a radio, were measured.
Gaussian noise is also known as the Gaussian distribution. In other words, the values that the noise can take on are Gaussian-distributed.
The strongest findings came when the so-called Gaussian noise was used, which involves a broad range of speeds and amplitudes. They believe that the noise emulates the feedback provided during slow sleep by biological neurons. The findings indicate that slow-wave sleep can in part help preserve cohesion, without hallucinating, of cortical neurons.
The next goal of the groups would be to use their Intel Loihi neuromorphic device algorithm. They hope Loihi can sleep occasionally so that information from a silicone retina cameras can be stabilized in real-time. If the results suggest that artificial brains need to sleep, we should possibly anticipate the same to happen for androids and other smart machines that could be created in the future.
Source: Los Alamos National Laboratory